2015-11-16
1(1)
Thesis proposal: Neural Networks for Part-of-Speech Tagging
Suitable for a Bachelor’s thesis (15 credits) or a project course
Background
Part-of-speech tagging is a central task in language technology. One simple but
effective technique for solving this task is based on classification, where the tag of a
word is predicted based on the word itself as well as context-based features such as
the next word, previous word, or tag of the previous word. A large variety of
machine learning models have been employed in this framework. One drawback of
many of these is their dependence on a large number of hand-crafted features.
Neural networks are an interesting alternative in this context because they are able
to learn features by themselves. The goal of this project is to explore the use of
neural networks for part-of-speech tagging.
Project description
Your task is to replace the machine learning components of an existing part-ofspeech tagger with a new model based on neural networks. To do so you will first
need to familiarise yourself with the existing literature on part-of-speech tagging
and neural networks. A significant part of the project will be devoted to
implementation work in Java or Python. After finishing the implementation, you
will evaluate the resulting system on standard data sets and compare it to existing
systems in the literature. At the end of the project you will prepare a scientific
report, documenting and discussing your findings. Depending on the outcome, the
project may be expanded into a Master’s thesis.
Student profile
You should be familiar with the basic principles of language technology (acquiring
and pre-processing textual data) and machine learning (linear classification, logistic
regression, standard neural network architectures). You should also be able to
program in either Java or Python.
Contact
Marco Kuhlmann, marco.kuhlmann@liu.se
LINKÖPING UNIVERSITY
DEPARTMENT OF COMPUTER AND INFORMATION SCIENCE/HUMAN-CENTRED SYSTEMS